def test_flop_with_output_shape(self):
inputs = [{
"image": torch.rand(3, 800, 800),
"height": 700,
"width": 700
}]
res = flop_count_operators(self.model, inputs)
self.assertEqual(int(res["conv"]), 117)
def test_flop(self):
# Faster R-CNN supports flop-counting with random inputs
inputs = [{"image": torch.rand(3, 800, 800)}]
res = flop_count_operators(self.model, inputs)
# This only checks flops for backbone & proposal generator
# Flops for box head is not conv, and depends on #proposals, which is
# almost 0 for random inputs.
self.assertTrue(int(res["conv"]), 117)
def do_flop(cfg):
data_loader = build_detection_test_loader(cfg, cfg.DATASETS.TEST[0])
model = build_model(cfg)
DetectionCheckpointer(model).load(cfg.MODEL.WEIGHTS)
model.eval()
counts = Counter()
for idx, data in zip(tqdm.trange(args.num_inputs), data_loader): # noqa
counts += flop_count_operators(model, data)
logger.info(
"(G)Flops for Each Type of Operators:\n" + str([(k, v / idx) for k, v in counts.items()])
)
def test_flop(self):
inputs1 = [{"image": torch.rand(3, 800, 800)}]
inputs2 = [{
"image": torch.rand(3, 800, 800),
"height": 700,
"width": 700
}]
for inputs in [inputs1, inputs2]:
res = flop_count_operators(self.model, inputs)
# The mask head could have extra conv flops, so total >= 117
self.assertGreaterEqual(int(res["conv"]), 117)
def do_flop(cfg):
data_loader = build_detection_test_loader(cfg, cfg.DATASETS.TEST[0])
model = build_model(cfg)
DetectionCheckpointer(model).load(cfg.MODEL.WEIGHTS)
model.eval()
counts = Counter()
total_flops = []
for idx, data in zip(tqdm.trange(args.num_inputs), data_loader): # noqa
count = flop_count_operators(model, data)
counts += count
total_flops.append(sum(count.values()))
logger.info("(G)Flops for Each Type of Operators:\n" +
str([(k, v / idx) for k, v in counts.items()]))
logger.info("Total (G)Flops: {}±{}".format(np.mean(total_flops),
np.std(total_flops)))
def test_flop(self):
# RetinaNet supports flop-counting with random inputs
inputs = [{"image": torch.rand(3, 800, 800)}]
res = flop_count_operators(self.model, inputs)
self.assertTrue(int(res["conv"]), 146) # 146B flops